As the decade wraps up, many of the most evident shifts in technology have already taken place in the AI, DL, and ML landscape. Billions in venture financing have been raised to chase AI opportunities, and media outlets now have dedicated beat reporters focused on the market. Though the space has come a long way, it’s also clear that there’s still a long way to go.
While reflecting on the past year, I started thinking about what 2020 may bring. I wanted to share some predictions on what will shape the industry landscape and the work I do at Determined AI.
Photo by: Photofusion/Universal Images Group via Getty Images
Exclusivity and Consolidation
In November 2019, we saw Graphcore make an exclusive deal with Microsoft - in many ways, the first of its kind. This deal provided Azure cloud customers exclusive access to Graphcore's Intelligence Processing Units (IPUs). Then in December we saw Intel buy Habana Labs, an AI chipmaker, for $2 billion to build out their AI strategy. I see this type of exclusive deal - between a legacy player and an AI startup - as a road to further adoption on a grander scale, and at a faster pace. However, I also think that this trend will continue to create clear winners & losers.
I believe that there will be a consolidation within the chip industry, and that several smaller AI chip companies will be bought by big players. The economics simply don’t make sense for most startups to manufacture chips.
It’s easy to see how these trends could lead to increased oligopoly pricing - or if we see one major player make the bulk of the moves, even monopoly pricing. If this pricing shift occurs, it means that AI chips will continue to be sold at a premium and only increase the haves versus have-nots in the AI space - continuing to widen the gap between players who are able to invest billions in an AI strategy, and those who simply never could.
The gap between the rest of the world and top AI companies - especially those focused on what we consider RedAI - will only widen. However, I think this divide will eventually lead the public to advocate for Green AI, focused on bringing awareness to the larger community and shaping important decisions for the next decade. This means greener research, greener algorithms, etc. in the future.
Big Growth for NVIDIA
In 2020, I believe NVIDIA’s data center revenue will continue to embrace double digit growth, quarter to quarter. Despite a couple of down quarters in early 2019, I see the rise of NVIDIA this year to be increasingly strong due to continued adoption of deep learning throughout the industry — partly fueled by more talent coming online and also new modeling approaches gaining adoption which I’ll touch more on in a subsequent prediction.
PyTorch > TensorFlow
Many advantages of PyTorch - from ramp up time to easier debugging - lead me to believe that PyTorch will continue to beat out TensorFlow, creating a systemic switch for teams and programmers in the long run. We’ll see more companies moving away from Google as developers advocate harder for PyTorch. Concretely, I predict that PyTorch will be the top framework (by the percentage of new projects it powers) by the end of the year.
I also think that we’ll see an increasing number of companies continue to invest in building out their AI teams. Stronger internal resources will lead companies to improve their on-premise delivery, therefore moving towards an increase in AI models and continued maturation in the space.
Deep Learning for Language Modeling Makes Big Jumps
Due to recent research advances (e.g. ELMo, Bert, the Transformer), I believe that within the next year, there will be massive adoption of deep learning for language modeling in industrial settings, everywhere that there is text. Within the year, we’ll see language modeling indistinguishable from humans at a level that most experts weren’t expecting for at least half a decade. Think something along the lines of a smartphone video input being translated to an AI-generated news story, indistinguishable from a journalist’s writing.
2020 Election
AI will play a critical - and likely negative - role in the upcoming U.S. presidential election. Deepfakes have the potential to further confuse an electorate already struggling with accusations of “fake news.” However, it is a game of cat and mouse, and some of the world’s most prominent researchers have been working on tools that can detect if images and videos have been retouched, altered, or digitally generated.
All in all, 2020 will no doubt bring exciting new innovations in the AI space. I am looking forward to what the next year will bring, and plan to continue to set my sights high for what’s coming next.
